Anima helps clinics cut call waits by 75% with GPT-6 and GPT-Live

Anima says it serves more than 14 million patients across over 1,000 UK clinics and processes more than 800,000 cases a month, and it's expanding to the U.S., Australia and beyond. It reports that phone waits at clinics using its voice receptionist fell from 40 to 50 minutes to about 10.

This is AI aimed at the administrative layer of medicine, the phone lines, letters and notes that pile up between a patient and a doctor. That layer is where delays turn into harm. The founder's own origin story is a patient whose abnormal results sat in a queue for six months.

We think this is the kind of healthcare AI worth rooting for. It isn't trying to diagnose. It's trying to move information faster to the people who make decisions. Anima's receptionist, Andi, runs on GPT-Live but routes intake through a proprietary policy engine with deterministic, clinically approved workflows. The model handles conversation; the rules stay fixed. That division of labor is exactly right for a clinical setting.

The document numbers are striking. Anima says practices that handled about 15 documents an hour now typically handle 30 to 45, and some clinics have cleared backlogs of more than 100,000 documents within days. If that holds, medication changes and test results reach clinicians in time instead of months later. That's not a productivity stat. That's patient safety.

The hard part is clinical coding. Anima's engineering lead notes there are around 800,000 possible SNOMED codes, and practices differ on which they use. Anima adapts with in-context learning rather than per-clinic fine-tuning. That's elegant, but it means a lot rides on the model following instructions consistently. A wrong code or a document matched to the wrong patient is a quiet error that can follow someone for years. The system suggests and staff review, and that review has to stay real, not ceremonial, as volumes climb.

On the scribe side, notes drafted in seconds and doctors finishing on time instead of working until 8 or 9 p.m. is a meaningful quality-of-life claim. Burnout is a clinical risk too. But automation bias is the predictable downside: a clinician who reviews hundreds of good drafts starts skimming the bad one.

We also note the 86% would-use-again figure and the observation that patients in their 80s and 90s were among the happiest users. Encouraging, and a useful rebuttal to the idea that older patients won't accept voice agents. The caveat for all of it: these are Anima's own reported figures in a vendor showcase. As it expands into the U.S., where privacy rules and payer complexity differ, we'd like to see independent outcome data.

Anima's platform combines Andi for phone intake, a Documents product built on GPT-6 models for extracting and routing clinical correspondence, and a Scribe product built on GPT 5.6 Luna for consultation notes, referral letters and follow-up messages. Anima says it chose OpenAI after benchmarking closed- and open-source models, and reports a median case resolution time of about 3.5 hours.